The European Commission has officially informed the EU Council of its concerns regarding the deadlines set for it for the follow-up measures provided for under the agreement concluded between the EU Council and the European Parliament on simplifying the rules on artificial intelligence, under the package known as the AI Omnibus (see EUROPE 13864/1). The EU Council is expected to give its final green light to the text on 29 June, following its approval by the European Parliament (see EUROPE 13889/23).
While welcoming the political agreement reached on 7 May on the AI Omnibus, the Commission considers the deadline set for the adoption of the delegated act — by 2 August 2027 — to be “problematic”, specifying whether, and under what conditions, certain requirements of the AI Act could be limited for high-risk AI systems where sectoral legislation already guarantees an equivalent or higher level of protection. It states that it “could limit its capacity to act if new overlaps were identified after the sectoral assessment provided for by that date has been carried out”. According to the Commission, “the value of this new mechanism lies precisely in its ability to address complementarity issues on an ongoing basis”. It stresses the need to improve consistency between the AI Act and sectoral legislation applicable to products. “Avoiding unnecessary duplication is essential to guarantee legal certainty, competitiveness and effective implementation”, it stresses in observations sent to the EU Council and seen by Agence Europe. The Commission also specifies that the agreement in no way prejudges its proposal to revise the Medical Devices Regulation.
As regards the guidelines it is to draw up on the interaction between the requirements applicable to high-risk AI systems and sectoral legislation by 1 August 2027, the Commission warns that a “thorough examination” will be necessary in order to ensure a “robust, practical and legally clear” outcome. Given that the application of the requirements relating to high-risk AI systems has been postponed until 2 August 2028, it considers that the obligation to adopt these guidelines “one year before those requirements become applicable could jeopardise the development of reliable support tools for industry and the authorities”.
For its part, Belgium also indicated that it still regarded the postponement of the obligation to label AI-generated content as “problematic”. Even if this postponement is justified by technical difficulties, it “raises major issues in terms of the protection of individuals and transparency regarding deepfakes that cannot be deferred without considerable risk”, it stresses.
It also regrets that the notion of “strict necessity” regarding bias detection was not clarified further and that recourse to alternative solutions, such as synthetic data, was not encouraged more strongly.
Lastly, regarding the agreement aimed at prohibiting AI systems capable of generating non-consensual sexually explicit content, Belgium underlines the importance of ensuring “adequate proactive protection in this area”, in full compliance with European values. (Original version in French by Ana Pisonero Hernández)